2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer.h,v 1.130 2008/11/13 02:18:43 dillon Exp $
37 * This header file contains structures used internally by the HAMMERFS
38 * implementation. See hammer_disk.h for on-disk structures.
41 #include <sys/param.h>
42 #include <sys/types.h>
43 #include <sys/kernel.h>
45 #include <sys/systm.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mountctl.h>
50 #include <sys/vnode.h>
54 #include <sys/globaldata.h>
55 #include <sys/lockf.h>
57 #include <sys/queue.h>
59 #include <sys/globaldata.h>
62 #include <sys/signal2.h>
63 #include "hammer_disk.h"
64 #include "hammer_mount.h"
65 #include "hammer_ioctl.h"
67 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
69 MALLOC_DECLARE(M_HAMMER);
74 #if !defined(KTR_HAMMER)
75 #define KTR_HAMMER KTR_ALL
77 KTR_INFO_MASTER_EXTERN(hammer);
85 * Key structure used for custom RB tree inode lookups. This prototypes
86 * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
88 typedef struct hammer_inode_info {
89 int64_t obj_id; /* (key) object identifier */
90 hammer_tid_t obj_asof; /* (key) snapshot transid or 0 */
91 u_int32_t obj_localization; /* (key) pseudo-fs */
93 struct hammer_btree_leaf_elm *leaf;
95 } *hammer_inode_info_t;
97 typedef enum hammer_transaction_type {
101 } hammer_transaction_type_t;
104 * HAMMER Transaction tracking
106 struct hammer_transaction {
107 hammer_transaction_type_t type;
108 struct hammer_mount *hmp;
114 struct hammer_volume *rootvol;
117 typedef struct hammer_transaction *hammer_transaction_t;
119 #define HAMMER_TRANSF_NEWINODE 0x0001
120 #define HAMMER_TRANSF_DIDIO 0x0002
121 #define HAMMER_TRANSF_CRCDOM 0x0004 /* EDOM on CRC error, less critical */
127 int refs; /* active references delay writes */
128 volatile u_int lockval; /* lock count and control bits */
129 struct thread *owner; /* owner if exclusively held */
132 #define HAMMER_LOCKF_EXCLUSIVE 0x40000000
133 #define HAMMER_LOCKF_WANTED 0x80000000
136 hammer_notlocked(struct hammer_lock *lock)
138 return(lock->lockval == 0);
142 hammer_islocked(struct hammer_lock *lock)
144 return(lock->lockval != 0);
148 hammer_isactive(struct hammer_lock *lock)
150 return(lock->refs != 0);
154 hammer_islastref(struct hammer_lock *lock)
156 return(lock->refs == 1);
160 * Return if we specifically own the lock exclusively.
163 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
165 if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
173 * Flush state, used by various structures
175 typedef enum hammer_inode_state {
179 } hammer_inode_state_t;
181 TAILQ_HEAD(hammer_record_list, hammer_record);
184 * Pseudo-filesystem extended data tracking
186 struct hammer_pfs_rb_tree;
187 struct hammer_pseudofs_inmem;
188 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
189 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
190 hammer_pfs_rb_compare, u_int32_t);
192 struct hammer_pseudofs_inmem {
193 RB_ENTRY(hammer_pseudofs_inmem) rb_node;
194 struct hammer_lock lock;
195 u_int32_t localization;
196 hammer_tid_t create_tid;
199 struct hammer_pseudofs_data pfsd;
202 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
204 #define HAMMER_PFSM_DELETED 0x0001
207 * Cache object ids. A fixed number of objid cache structures are
208 * created to reserve object id's for newly created files in multiples
209 * of 100,000, localized to a particular directory, and recycled as
210 * needed. This allows parallel create operations in different
211 * directories to retain fairly localized object ids which in turn
212 * improves reblocking performance and layout.
214 #define OBJID_CACHE_SIZE 1024
215 #define OBJID_CACHE_BULK_BITS 10 /* 10 bits (1024) */
216 #define OBJID_CACHE_BULK (32 * 32) /* two level (1024) */
217 #define OBJID_CACHE_BULK_MASK (OBJID_CACHE_BULK - 1)
218 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
220 typedef struct hammer_objid_cache {
221 TAILQ_ENTRY(hammer_objid_cache) entry;
222 struct hammer_inode *dip;
223 hammer_tid_t base_tid;
227 } *hammer_objid_cache_t;
230 * Associate an inode with a B-Tree node to cache search start positions
232 typedef struct hammer_node_cache {
233 TAILQ_ENTRY(hammer_node_cache) entry;
234 struct hammer_node *node;
235 struct hammer_inode *ip;
236 } *hammer_node_cache_t;
238 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
241 * Structure used to organize flush groups. Flush groups must be
242 * organized into chunks in order to avoid blowing out the UNDO FIFO.
243 * Without this a 'sync' could end up flushing 50,000 inodes in a single
246 struct hammer_flush_group {
247 TAILQ_ENTRY(hammer_flush_group) flush_entry;
248 TAILQ_HEAD(, hammer_inode) flush_list;
249 int unused01; /* inode load */
250 int total_count; /* record load */
251 int running; /* group is running */
256 typedef struct hammer_flush_group *hammer_flush_group_t;
258 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
261 * Structure used to represent an inode in-memory.
263 * The record and data associated with an inode may be out of sync with
264 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
267 * An inode may also hold a cache of unsynchronized records, used for
268 * database and directories only. Unsynchronized regular file data is
269 * stored in the buffer cache.
271 * NOTE: A file which is created and destroyed within the initial
272 * synchronization period can wind up not doing any disk I/O at all.
274 * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
276 struct hammer_ino_rb_tree;
278 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
279 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
280 hammer_ino_rb_compare, hammer_inode_info_t);
282 struct hammer_rec_rb_tree;
283 struct hammer_record;
284 RB_HEAD(hammer_rec_rb_tree, hammer_record);
285 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
286 hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
288 TAILQ_HEAD(hammer_node_list, hammer_node);
290 struct hammer_inode {
291 RB_ENTRY(hammer_inode) rb_node;
292 hammer_inode_state_t flush_state;
293 hammer_flush_group_t flush_group;
294 TAILQ_ENTRY(hammer_inode) flush_entry;
295 struct hammer_record_list target_list; /* target of dependant recs */
296 int64_t obj_id; /* (key) object identifier */
297 hammer_tid_t obj_asof; /* (key) snapshot or 0 */
298 u_int32_t obj_localization; /* (key) pseudo-fs */
299 struct hammer_mount *hmp;
300 hammer_objid_cache_t objid_cache;
302 int error; /* flush error */
303 int cursor_ip_refs; /* sanity */
306 hammer_pseudofs_inmem_t pfsm;
307 struct lockf advlock;
308 struct hammer_lock lock; /* sync copy interlock */
310 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */
311 struct hammer_inode_data ino_data; /* in-memory cache */
312 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */
314 struct hammer_node_cache cache[4]; /* search initiate cache */
317 * When a demark is created to synchronize an inode to
318 * disk, certain fields are copied so the front-end VOPs
319 * can continue to run in parallel with the synchronization
320 * occuring in the background.
322 int sync_flags; /* to-sync flags cache */
323 off_t sync_trunc_off; /* to-sync truncation */
324 off_t save_trunc_off; /* write optimization */
325 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
326 struct hammer_inode_data sync_ino_data; /* to-sync cache */
329 typedef struct hammer_inode *hammer_inode_t;
331 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data)
333 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */
334 /* (not including atime/mtime) */
335 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */
336 #define HAMMER_INODE_CONN_DOWN 0x0004 /* include in downward recursion */
337 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */
338 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */
339 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */
340 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */
341 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */
342 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */
343 #define HAMMER_INODE_VHELD 0x0400 /* vnode held on sync */
344 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */
345 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */
346 #define HAMMER_INODE_REFLUSH 0x2000 /* flush on dependancy / reflush */
347 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */
348 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */
350 #define HAMMER_INODE_TRUNCATED 0x00010000
351 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/
352 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */
353 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */
354 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */
355 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
356 #define HAMMER_INODE_DUMMY 0x00800000 /* dummy inode covering bad file */
358 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY| \
359 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \
360 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \
361 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
363 #define HAMMER_INODE_MODMASK_NOXDIRTY \
364 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
366 #define HAMMER_FLUSH_GROUP_SIZE 64
368 #define HAMMER_FLUSH_SIGNAL 0x0001
369 #define HAMMER_FLUSH_RECURSION 0x0002
372 * Used by the inode reclaim code to pipeline reclaims and avoid
373 * blowing out kernel memory or letting the flusher get too far
374 * behind. The reclaim wakes up when count reaches 0 or the
377 struct hammer_reclaim {
378 TAILQ_ENTRY(hammer_reclaim) entry;
382 #define HAMMER_RECLAIM_FLUSH 2000
383 #define HAMMER_RECLAIM_WAIT 4000
386 * Structure used to represent an unsynchronized record in-memory. These
387 * records typically represent directory entries. Only non-historical
388 * records are kept in-memory.
390 * Records are organized as a per-inode RB-Tree. If the inode is not
391 * on disk then neither are any records and the in-memory record tree
392 * represents the entire contents of the inode. If the inode is on disk
393 * then the on-disk B-Tree is scanned in parallel with the in-memory
394 * RB-Tree to synthesize the current state of the file.
396 * Records are also used to enforce the ordering of directory create/delete
397 * operations. A new inode will not be flushed to disk unless its related
398 * directory entry is also being flushed at the same time. A directory entry
399 * will not be removed unless its related inode is also being removed at the
402 typedef enum hammer_record_type {
403 HAMMER_MEM_RECORD_GENERAL, /* misc record */
404 HAMMER_MEM_RECORD_INODE, /* inode record */
405 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */
406 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */
407 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */
408 } hammer_record_type_t;
410 struct hammer_record {
411 RB_ENTRY(hammer_record) rb_node;
412 TAILQ_ENTRY(hammer_record) target_entry;
413 hammer_inode_state_t flush_state;
414 hammer_flush_group_t flush_group;
415 hammer_record_type_t type;
416 struct hammer_lock lock;
417 struct hammer_reserve *resv;
418 struct hammer_inode *ip;
419 struct hammer_inode *target_ip;
420 struct hammer_btree_leaf_elm leaf;
421 union hammer_data_ondisk *data;
423 hammer_off_t zone2_offset; /* direct-write only */
426 typedef struct hammer_record *hammer_record_t;
429 * Record flags. Note that FE can only be set by the frontend if the
430 * record has not been interlocked by the backend w/ BE.
432 #define HAMMER_RECF_ALLOCDATA 0x0001
433 #define HAMMER_RECF_ONRBTREE 0x0002
434 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */
435 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */
436 #define HAMMER_RECF_COMMITTED 0x0010 /* committed to the B-Tree */
437 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */
438 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */
439 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */
440 #define HAMMER_RECF_DIRECT_IO 0x0200 /* related direct I/O running*/
441 #define HAMMER_RECF_DIRECT_WAIT 0x0400 /* related direct I/O running*/
442 #define HAMMER_RECF_DIRECT_INVAL 0x0800 /* buffer alias invalidation */
445 * hammer_delete_at_cursor() flags
447 #define HAMMER_DELETE_ADJUST 0x0001
448 #define HAMMER_DELETE_DESTROY 0x0002
451 * In-memory structures representing on-disk structures.
453 struct hammer_volume;
454 struct hammer_buffer;
457 struct hammer_reserve;
459 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
460 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
461 RB_HEAD(hammer_nod_rb_tree, hammer_node);
462 RB_HEAD(hammer_und_rb_tree, hammer_undo);
463 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
465 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
466 hammer_vol_rb_compare, int32_t);
467 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
468 hammer_buf_rb_compare, hammer_off_t);
469 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
470 hammer_nod_rb_compare, hammer_off_t);
471 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
472 hammer_und_rb_compare, hammer_off_t);
473 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
474 hammer_res_rb_compare, hammer_off_t);
477 * IO management - embedded at the head of various in-memory structures
479 * VOLUME - hammer_volume containing meta-data
480 * META_BUFFER - hammer_buffer containing meta-data
481 * DATA_BUFFER - hammer_buffer containing pure-data
483 * Dirty volume headers and dirty meta-data buffers are locked until the
484 * flusher can sequence them out. Dirty pure-data buffers can be written.
485 * Clean buffers can be passively released.
487 typedef enum hammer_io_type {
488 HAMMER_STRUCTURE_VOLUME,
489 HAMMER_STRUCTURE_META_BUFFER,
490 HAMMER_STRUCTURE_UNDO_BUFFER,
491 HAMMER_STRUCTURE_DATA_BUFFER
494 union hammer_io_structure;
498 LIST_ENTRY(worklist) node;
501 TAILQ_HEAD(hammer_io_list, hammer_io);
502 typedef struct hammer_io_list *hammer_io_list_t;
505 struct worklist worklist;
506 struct hammer_lock lock;
507 enum hammer_io_type type;
508 struct hammer_mount *hmp;
509 struct hammer_volume *volume;
510 TAILQ_ENTRY(hammer_io) mod_entry; /* list entry if modified */
511 hammer_io_list_t mod_list;
513 int64_t offset; /* zone-2 offset */
514 int bytes; /* buffer cache buffer size */
515 int loading; /* loading/unloading interlock */
518 u_int modified : 1; /* bp's data was modified */
519 u_int released : 1; /* bp released (w/ B_LOCKED set) */
520 u_int running : 1; /* bp write IO in progress */
521 u_int waiting : 1; /* someone is waiting on us */
522 u_int validated : 1; /* ondisk has been validated */
523 u_int waitdep : 1; /* flush waits for dependancies */
524 u_int recovered : 1; /* has recovery ref */
525 u_int waitmod : 1; /* waiting for modify_refs */
526 u_int reclaim : 1; /* reclaim requested */
527 u_int gencrc : 1; /* crc needs to be generated */
528 u_int ioerror : 1; /* abort on io-error */
531 typedef struct hammer_io *hammer_io_t;
533 #define HAMMER_CLUSTER_SIZE (64 * 1024)
534 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
535 #undef HAMMER_CLUSTER_SIZE
536 #define HAMMER_CLUSTER_SIZE MAXBSIZE
538 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
541 * In-memory volume representing on-disk buffer
543 struct hammer_volume {
545 RB_ENTRY(hammer_volume) rb_node;
546 struct hammer_volume_ondisk *ondisk;
548 int64_t nblocks; /* note: special calculation for statfs */
549 int64_t buffer_base; /* base offset of buffer 0 */
550 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
551 hammer_off_t maxraw_off; /* Maximum raw offset for device */
557 typedef struct hammer_volume *hammer_volume_t;
560 * In-memory buffer (other then volume, super-cluster, or cluster),
561 * representing an on-disk buffer.
563 struct hammer_buffer {
565 RB_ENTRY(hammer_buffer) rb_node;
567 hammer_off_t zoneX_offset;
568 hammer_off_t zone2_offset;
569 struct hammer_reserve *resv;
570 struct hammer_node_list clist;
573 typedef struct hammer_buffer *hammer_buffer_t;
576 * In-memory B-Tree node, representing an on-disk B-Tree node.
578 * This is a hang-on structure which is backed by a hammer_buffer,
579 * indexed by a hammer_cluster, and used for fine-grained locking of
580 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer
581 * can contain multiple nodes representing wildly disassociated portions
582 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
584 * This structure uses a cluster-relative index to reduce the number
585 * of layers required to access it, and also because all on-disk B-Tree
586 * references are cluster-relative offsets.
589 struct hammer_lock lock; /* node-by-node lock */
590 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */
591 RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */
592 hammer_off_t node_offset; /* full offset spec */
593 struct hammer_mount *hmp;
594 struct hammer_buffer *buffer; /* backing buffer */
595 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */
596 TAILQ_HEAD(, hammer_cursor) cursor_list; /* deadlock recovery */
597 struct hammer_node_cache_list cache_list; /* passive caches */
599 int loading; /* load interlock */
602 #define HAMMER_NODE_DELETED 0x0001
603 #define HAMMER_NODE_FLUSH 0x0002
604 #define HAMMER_NODE_CRCGOOD 0x0004
605 #define HAMMER_NODE_NEEDSCRC 0x0008
606 #define HAMMER_NODE_NEEDSMIRROR 0x0010
607 #define HAMMER_NODE_CRCBAD 0x0020
609 #define HAMMER_NODE_CRCANY (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
611 typedef struct hammer_node *hammer_node_t;
614 * List of locked nodes. This structure is used to lock potentially large
615 * numbers of nodes as an aid for complex B-Tree operations.
617 struct hammer_node_lock;
618 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
620 struct hammer_node_lock {
621 TAILQ_ENTRY(hammer_node_lock) entry;
622 struct hammer_node_lock_list list;
623 struct hammer_node_lock *parent;
625 hammer_node_ondisk_t copy; /* copy of on-disk data */
626 int index; /* index of this node in parent */
627 int count; /* count children */
631 typedef struct hammer_node_lock *hammer_node_lock_t;
633 #define HAMMER_NODE_LOCK_UPDATED 0x0001
636 * Common I/O management structure - embedded in in-memory structures
637 * which are backed by filesystem buffers.
639 union hammer_io_structure {
641 struct hammer_volume volume;
642 struct hammer_buffer buffer;
645 typedef union hammer_io_structure *hammer_io_structure_t;
648 * The reserve structure prevents the blockmap from allocating
649 * out of a reserved bigblock. Such reservations are used by
650 * the direct-write mechanism.
652 * The structure is also used to hold off on reallocations of
653 * big blocks from the freemap until flush dependancies have
656 struct hammer_reserve {
657 RB_ENTRY(hammer_reserve) rb_node;
658 TAILQ_ENTRY(hammer_reserve) delay_entry;
664 hammer_off_t zone_offset;
667 typedef struct hammer_reserve *hammer_reserve_t;
669 #define HAMMER_RESF_ONDELAY 0x0001
670 #define HAMMER_RESF_LAYER2FREE 0x0002
672 #include "hammer_cursor.h"
675 * The undo structure tracks recent undos to avoid laying down duplicate
676 * undos within a flush group, saving us a significant amount of overhead.
678 * This is strictly a heuristic.
680 #define HAMMER_MAX_UNDOS 1024
681 #define HAMMER_MAX_FLUSHERS 4
684 RB_ENTRY(hammer_undo) rb_node;
685 TAILQ_ENTRY(hammer_undo) lru_entry;
690 typedef struct hammer_undo *hammer_undo_t;
692 struct hammer_flusher_info;
693 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
695 struct hammer_flusher {
696 int signal; /* flusher thread sequencer */
697 int act; /* currently active flush group */
698 int done; /* set to act when complete */
699 int next; /* next flush group */
700 int group_lock; /* lock sequencing of the next flush */
701 int exiting; /* request master exit */
702 thread_t td; /* master flusher thread */
703 hammer_tid_t tid; /* last flushed transaction id */
704 int finalize_want; /* serialize finalization */
705 struct hammer_lock finalize_lock; /* serialize finalization */
706 struct hammer_transaction trans; /* shared transaction */
707 struct hammer_flusher_info_list run_list;
708 struct hammer_flusher_info_list ready_list;
712 * Internal hammer mount data structure
714 struct hammer_mount {
716 /*struct vnode *rootvp;*/
717 struct hammer_ino_rb_tree rb_inos_root;
718 struct hammer_vol_rb_tree rb_vols_root;
719 struct hammer_nod_rb_tree rb_nods_root;
720 struct hammer_und_rb_tree rb_undo_root;
721 struct hammer_res_rb_tree rb_resv_root;
722 struct hammer_buf_rb_tree rb_bufs_root;
723 struct hammer_pfs_rb_tree rb_pfsm_root;
724 struct hammer_volume *rootvol;
725 struct hammer_base_elm root_btree_beg;
726 struct hammer_base_elm root_btree_end;
728 struct malloc_type *m_misc;
729 struct malloc_type *m_inodes;
731 int flags; /* HAMMER_MOUNT_xxx flags */
736 int master_id; /* -1 or 0-15 - clustering and mirroring */
737 int version; /* hammer filesystem version to use */
738 int rsv_inodes; /* reserved space due to dirty inodes */
739 int64_t rsv_databytes; /* reserved space due to record data */
740 int rsv_recs; /* reserved space due to dirty records */
741 int rsv_fromdelay; /* bigblocks reserved due to flush delay */
742 int undo_rec_limit; /* based on size of undo area */
744 int count_newrecords;
746 int inode_reclaims; /* inodes pending reclaim by flusher */
747 int count_inodes; /* total number of inodes */
748 int count_iqueued; /* inodes queued to flusher */
750 struct hammer_flusher flusher;
752 u_int check_interrupt;
754 struct hammer_io_list volu_list; /* dirty undo buffers */
755 struct hammer_io_list undo_list; /* dirty undo buffers */
756 struct hammer_io_list data_list; /* dirty data buffers */
757 struct hammer_io_list alt_data_list; /* dirty data buffers */
758 struct hammer_io_list meta_list; /* dirty meta bufs */
759 struct hammer_io_list lose_list; /* loose buffers */
760 int locked_dirty_space; /* meta/volu count */
761 int io_running_space;
762 int objid_cache_count;
763 int error; /* critical I/O error */
764 struct krate krate; /* rate limited kprintf */
765 hammer_tid_t asof; /* snapshot mount */
766 hammer_tid_t next_tid;
767 hammer_tid_t flush_tid1; /* flusher tid sequencing */
768 hammer_tid_t flush_tid2; /* flusher tid sequencing */
769 int64_t copy_stat_freebigblocks; /* number of free bigblocks */
771 struct netexport export;
772 struct hammer_lock sync_lock;
773 struct hammer_lock free_lock;
774 struct hammer_lock undo_lock;
775 struct hammer_lock blkmap_lock;
776 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES];
777 struct hammer_undo undos[HAMMER_MAX_UNDOS];
779 TAILQ_HEAD(, hammer_undo) undo_lru_list;
780 TAILQ_HEAD(, hammer_reserve) delay_list;
781 struct hammer_flush_group_list flush_group_list;
782 hammer_flush_group_t next_flush_group;
783 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
784 TAILQ_HEAD(, hammer_reclaim) reclaim_list;
787 typedef struct hammer_mount *hammer_mount_t;
789 #define HAMMER_MOUNT_CRITICAL_ERROR 0x0001
790 #define HAMMER_MOUNT_FLUSH_RECOVERY 0x0002
792 struct hammer_sync_info {
800 * checkspace slop (8MB chunks), higher numbers are more conservative.
802 #define HAMMER_CHKSPC_REBLOCK 25
803 #define HAMMER_CHKSPC_MIRROR 20
804 #define HAMMER_CHKSPC_WRITE 20
805 #define HAMMER_CHKSPC_CREATE 20
806 #define HAMMER_CHKSPC_REMOVE 10
807 #define HAMMER_CHKSPC_EMERGENCY 0
811 extern struct vop_ops hammer_vnode_vops;
812 extern struct vop_ops hammer_spec_vops;
813 extern struct vop_ops hammer_fifo_vops;
814 extern struct bio_ops hammer_bioops;
816 extern int hammer_debug_io;
817 extern int hammer_debug_general;
818 extern int hammer_debug_debug;
819 extern int hammer_debug_inode;
820 extern int hammer_debug_locks;
821 extern int hammer_debug_btree;
822 extern int hammer_debug_tid;
823 extern int hammer_debug_recover;
824 extern int hammer_debug_recover_faults;
825 extern int hammer_cluster_enable;
826 extern int hammer_count_fsyncs;
827 extern int hammer_count_inodes;
828 extern int hammer_count_iqueued;
829 extern int hammer_count_reclaiming;
830 extern int hammer_count_records;
831 extern int hammer_count_record_datas;
832 extern int hammer_count_volumes;
833 extern int hammer_count_buffers;
834 extern int hammer_count_nodes;
835 extern int64_t hammer_count_extra_space_used;
836 extern int64_t hammer_stats_btree_lookups;
837 extern int64_t hammer_stats_btree_searches;
838 extern int64_t hammer_stats_btree_inserts;
839 extern int64_t hammer_stats_btree_deletes;
840 extern int64_t hammer_stats_btree_elements;
841 extern int64_t hammer_stats_btree_splits;
842 extern int64_t hammer_stats_btree_iterations;
843 extern int64_t hammer_stats_btree_root_iterations;
844 extern int64_t hammer_stats_record_iterations;
845 extern int64_t hammer_stats_file_read;
846 extern int64_t hammer_stats_file_write;
847 extern int64_t hammer_stats_file_iopsr;
848 extern int64_t hammer_stats_file_iopsw;
849 extern int64_t hammer_stats_disk_read;
850 extern int64_t hammer_stats_disk_write;
851 extern int64_t hammer_stats_inode_flushes;
852 extern int64_t hammer_stats_commits;
853 extern int hammer_count_dirtybufspace;
854 extern int hammer_count_refedbufs;
855 extern int hammer_count_reservations;
856 extern int hammer_count_io_running_read;
857 extern int hammer_count_io_running_write;
858 extern int hammer_count_io_locked;
859 extern int hammer_limit_dirtybufspace;
860 extern int hammer_limit_recs;
861 extern int hammer_bio_count;
862 extern int hammer_verify_zone;
863 extern int hammer_verify_data;
864 extern int hammer_write_mode;
865 extern int hammer_autoflush;
866 extern int64_t hammer_contention_count;
868 void hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
869 int error, const char *msg);
870 int hammer_vop_inactive(struct vop_inactive_args *);
871 int hammer_vop_reclaim(struct vop_reclaim_args *);
872 int hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
873 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
874 hammer_inode_t dip, int64_t obj_id,
875 hammer_tid_t asof, u_int32_t localization,
876 int flags, int *errorp);
877 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
878 hammer_inode_t dip, int64_t obj_id,
879 hammer_tid_t asof, u_int32_t localization,
880 int flags, int *errorp);
881 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
882 int64_t obj_id, hammer_tid_t asof,
883 u_int32_t localization);
884 void hammer_scan_inode_snapshots(hammer_mount_t hmp,
885 hammer_inode_info_t iinfo,
886 int (*callback)(hammer_inode_t ip, void *data),
888 void hammer_put_inode(struct hammer_inode *ip);
889 void hammer_put_inode_ref(struct hammer_inode *ip);
890 void hammer_inode_waitreclaims(hammer_mount_t hmp);
891 void hammer_inode_waithard(hammer_mount_t hmp);
893 int hammer_unload_volume(hammer_volume_t volume, void *data __unused);
894 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
896 int hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused);
897 int hammer_install_volume(hammer_mount_t hmp, const char *volname,
898 struct vnode *devvp);
899 int hammer_mountcheck_volumes(hammer_mount_t hmp);
901 int hammer_mem_add(hammer_record_t record);
902 int hammer_ip_lookup(hammer_cursor_t cursor);
903 int hammer_ip_first(hammer_cursor_t cursor);
904 int hammer_ip_next(hammer_cursor_t cursor);
905 int hammer_ip_resolve_data(hammer_cursor_t cursor);
906 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
908 int hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
909 hammer_tid_t delete_tid, u_int32_t delete_ts,
910 int track, int64_t *stat_bytes);
911 int hammer_ip_check_directory_empty(hammer_transaction_t trans,
913 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
914 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
917 hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
918 void hammer_flush_record_done(hammer_record_t record, int error);
919 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
920 void hammer_rel_mem_record(hammer_record_t record);
922 int hammer_cursor_up(hammer_cursor_t cursor);
923 int hammer_cursor_up_locked(hammer_cursor_t cursor);
924 int hammer_cursor_down(hammer_cursor_t cursor);
925 int hammer_cursor_upgrade(hammer_cursor_t cursor);
926 int hammer_cursor_upgrade_node(hammer_cursor_t cursor);
927 void hammer_cursor_downgrade(hammer_cursor_t cursor);
928 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
930 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
931 int hammer_lock_ex_try(struct hammer_lock *lock);
932 void hammer_lock_sh(struct hammer_lock *lock);
933 int hammer_lock_sh_try(struct hammer_lock *lock);
934 int hammer_lock_upgrade(struct hammer_lock *lock);
935 void hammer_lock_downgrade(struct hammer_lock *lock);
936 int hammer_lock_status(struct hammer_lock *lock);
937 void hammer_unlock(struct hammer_lock *lock);
938 void hammer_ref(struct hammer_lock *lock);
939 void hammer_unref(struct hammer_lock *lock);
941 void hammer_sync_lock_ex(hammer_transaction_t trans);
942 void hammer_sync_lock_sh(hammer_transaction_t trans);
943 int hammer_sync_lock_sh_try(hammer_transaction_t trans);
944 void hammer_sync_unlock(hammer_transaction_t trans);
946 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
947 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
948 void hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
949 u_int64_t hammer_timespec_to_time(struct timespec *ts);
950 int hammer_str_to_tid(const char *str, int *ispfsp,
951 hammer_tid_t *tidp, u_int32_t *localizationp);
952 int hammer_is_atatext(const char *name, int len);
953 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
955 void hammer_clear_objid(hammer_inode_t dip);
956 void hammer_destroy_objid_cache(hammer_mount_t hmp);
958 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
960 void hammer_clear_undo_history(hammer_mount_t hmp);
961 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
962 int hammer_get_dtype(u_int8_t obj_type);
963 u_int8_t hammer_get_obj_type(enum vtype vtype);
964 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
965 u_int32_t *max_iterationsp);
966 int hammer_nohistory(hammer_inode_t ip);
968 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
969 hammer_node_cache_t cache, hammer_inode_t ip);
970 void hammer_normalize_cursor(hammer_cursor_t cursor);
971 void hammer_done_cursor(hammer_cursor_t cursor);
972 int hammer_recover_cursor(hammer_cursor_t cursor);
973 void hammer_unlock_cursor(hammer_cursor_t cursor);
974 int hammer_lock_cursor(hammer_cursor_t cursor);
975 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
976 void hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
978 void hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
979 void hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
981 void hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
983 void hammer_cursor_moved_element(hammer_node_t onode, hammer_node_t nnode,
984 int oindex, int nindex);
985 void hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
986 hammer_node_t nparent, int nindex);
987 void hammer_cursor_inserted_element(hammer_node_t node, int index);
988 void hammer_cursor_deleted_element(hammer_node_t node, int index);
990 int hammer_btree_lookup(hammer_cursor_t cursor);
991 int hammer_btree_first(hammer_cursor_t cursor);
992 int hammer_btree_last(hammer_cursor_t cursor);
993 int hammer_btree_extract(hammer_cursor_t cursor, int flags);
994 int hammer_btree_iterate(hammer_cursor_t cursor);
995 int hammer_btree_iterate_reverse(hammer_cursor_t cursor);
996 int hammer_btree_insert(hammer_cursor_t cursor,
997 hammer_btree_leaf_elm_t elm, int *doprop);
998 int hammer_btree_delete(hammer_cursor_t cursor);
999 void hammer_btree_do_propagation(hammer_cursor_t cursor,
1000 hammer_pseudofs_inmem_t pfsm,
1001 hammer_btree_leaf_elm_t leaf);
1002 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1003 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1004 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1005 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1007 int btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
1008 hammer_btree_elm_t elm);
1009 void hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1010 int hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1011 hammer_node_lock_t parent);
1012 void hammer_btree_lock_copy(hammer_cursor_t cursor,
1013 hammer_node_lock_t parent);
1014 int hammer_btree_sync_copy(hammer_cursor_t cursor,
1015 hammer_node_lock_t parent);
1016 void hammer_btree_unlock_children(hammer_cursor_t cursor,
1017 hammer_node_lock_t parent);
1018 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1019 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1020 hammer_node_t node, int *parent_indexp,
1021 int *errorp, int try_exclusive);
1023 void hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1024 void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
1026 void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
1027 int *errorp, struct hammer_buffer **bufferp);
1028 void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
1029 int *errorp, struct hammer_buffer **bufferp);
1030 void *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1031 int *errorp, struct hammer_buffer **bufferp);
1032 void *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1033 int *errorp, struct hammer_buffer **bufferp);
1035 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1037 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1038 int32_t vol_no, int *errorp);
1039 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1040 int bytes, int isnew, int *errorp);
1041 void hammer_sync_buffers(hammer_mount_t hmp,
1042 hammer_off_t base_offset, int bytes);
1043 int hammer_del_buffers(hammer_mount_t hmp,
1044 hammer_off_t base_offset,
1045 hammer_off_t zone2_offset, int bytes,
1046 int report_conflicts);
1048 int hammer_ref_volume(hammer_volume_t volume);
1049 int hammer_ref_buffer(hammer_buffer_t buffer);
1050 void hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1052 void hammer_rel_volume(hammer_volume_t volume, int flush);
1053 void hammer_rel_buffer(hammer_buffer_t buffer, int flush);
1055 int hammer_vfs_export(struct mount *mp, int op,
1056 const struct export_args *export);
1057 hammer_node_t hammer_get_node(hammer_transaction_t trans,
1058 hammer_off_t node_offset, int isnew, int *errorp);
1059 void hammer_ref_node(hammer_node_t node);
1060 hammer_node_t hammer_ref_node_safe(hammer_transaction_t trans,
1061 hammer_node_cache_t cache, int *errorp);
1062 void hammer_rel_node(hammer_node_t node);
1063 void hammer_delete_node(hammer_transaction_t trans,
1064 hammer_node_t node);
1065 void hammer_cache_node(hammer_node_cache_t cache,
1066 hammer_node_t node);
1067 void hammer_uncache_node(hammer_node_cache_t cache);
1068 void hammer_flush_node(hammer_node_t node);
1070 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1071 struct hammer_buffer *buffer);
1072 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1073 hammer_off_t hint, int *errorp);
1074 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1075 u_int16_t rec_type, hammer_off_t *data_offsetp,
1076 struct hammer_buffer **data_bufferp,
1077 hammer_off_t hint, int *errorp);
1079 int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io,
1080 hammer_off_t zone1_offset, void *base, int len);
1082 void hammer_put_volume(struct hammer_volume *volume, int flush);
1083 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1085 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1086 hammer_off_t owner, int *errorp);
1087 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1088 hammer_off_t owner, int *errorp);
1089 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1090 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1091 int bytes, hammer_off_t hint, int *errorp);
1092 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1093 int bytes, hammer_off_t *zone_offp, int *errorp);
1094 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1095 hammer_reserve_t resv);
1096 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1097 void hammer_blockmap_free(hammer_transaction_t trans,
1098 hammer_off_t bmap_off, int bytes);
1099 int hammer_blockmap_finalize(hammer_transaction_t trans,
1100 hammer_reserve_t resv,
1101 hammer_off_t bmap_off, int bytes);
1102 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1103 int *curp, int *errorp);
1104 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1106 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1108 int64_t hammer_undo_used(hammer_transaction_t trans);
1109 int64_t hammer_undo_space(hammer_transaction_t trans);
1110 int64_t hammer_undo_max(hammer_mount_t hmp);
1112 void hammer_start_transaction(struct hammer_transaction *trans,
1113 struct hammer_mount *hmp);
1114 void hammer_simple_transaction(struct hammer_transaction *trans,
1115 struct hammer_mount *hmp);
1116 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1117 struct hammer_mount *hmp);
1118 void hammer_done_transaction(struct hammer_transaction *trans);
1120 void hammer_modify_inode(hammer_inode_t ip, int flags);
1121 void hammer_flush_inode(hammer_inode_t ip, int flags);
1122 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1123 void hammer_wait_inode(hammer_inode_t ip);
1125 int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1126 struct ucred *cred, struct hammer_inode *dip,
1127 const char *name, int namelen,
1128 hammer_pseudofs_inmem_t pfsm,
1129 struct hammer_inode **ipp);
1130 void hammer_rel_inode(hammer_inode_t ip, int flush);
1131 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1132 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1133 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1135 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1136 void hammer_test_inode(hammer_inode_t dip);
1137 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1139 int hammer_ip_add_directory(struct hammer_transaction *trans,
1140 hammer_inode_t dip, const char *name, int bytes,
1141 hammer_inode_t nip);
1142 int hammer_ip_del_directory(struct hammer_transaction *trans,
1143 hammer_cursor_t cursor, hammer_inode_t dip,
1145 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1146 void *data, int bytes, int *errorp);
1147 int hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1148 int hammer_ip_add_record(struct hammer_transaction *trans,
1149 hammer_record_t record);
1150 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1151 int64_t ran_beg, int64_t ran_end, int truncating);
1152 int hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1154 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1155 int64_t offset, void *data, int bytes);
1156 int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1157 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1158 hammer_pseudofs_inmem_t hammer_load_pseudofs(hammer_transaction_t trans,
1159 u_int32_t localization, int *errorp);
1160 int hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1161 hammer_pseudofs_inmem_t pfsm);
1162 int hammer_save_pseudofs(hammer_transaction_t trans,
1163 hammer_pseudofs_inmem_t pfsm);
1164 int hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1165 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1166 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1167 struct ucred *cred);
1169 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1170 enum hammer_io_type type);
1171 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
1172 hammer_off_t limit);
1173 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1174 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1175 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1176 void hammer_io_flush(struct hammer_io *io);
1177 void hammer_io_wait(struct hammer_io *io);
1178 void hammer_io_waitdep(struct hammer_io *io);
1179 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident);
1180 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1181 hammer_btree_leaf_elm_t leaf);
1182 int hammer_io_direct_write(hammer_mount_t hmp, hammer_record_t record,
1184 void hammer_io_direct_wait(hammer_record_t record);
1185 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1186 void hammer_io_write_interlock(hammer_io_t io);
1187 void hammer_io_done_interlock(hammer_io_t io);
1188 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1189 void hammer_io_clear_modlist(struct hammer_io *io);
1190 void hammer_io_flush_sync(hammer_mount_t hmp);
1192 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1193 void *base, int len);
1194 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1195 void *base, int len);
1196 void hammer_modify_volume_done(hammer_volume_t volume);
1197 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1199 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1200 struct hammer_ioc_reblock *reblock);
1201 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1202 struct hammer_ioc_rebalance *rebal);
1203 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1204 struct hammer_ioc_prune *prune);
1205 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1206 struct hammer_ioc_mirror_rw *mirror);
1207 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1208 struct hammer_ioc_mirror_rw *mirror);
1209 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1210 struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1211 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1212 struct hammer_ioc_pseudofs_rw *pfs);
1213 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1214 struct hammer_ioc_pseudofs_rw *pfs);
1215 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1216 struct hammer_ioc_pseudofs_rw *pfs);
1217 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1218 struct hammer_ioc_pseudofs_rw *pfs);
1219 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1220 struct hammer_ioc_pseudofs_rw *pfs);
1221 int hammer_ioc_expand(hammer_transaction_t trans, hammer_inode_t ip,
1222 struct hammer_ioc_expand *expand);
1224 int hammer_signal_check(hammer_mount_t hmp);
1226 void hammer_flusher_create(hammer_mount_t hmp);
1227 void hammer_flusher_destroy(hammer_mount_t hmp);
1228 void hammer_flusher_sync(hammer_mount_t hmp);
1229 int hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1230 int hammer_flusher_async_one(hammer_mount_t hmp);
1231 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1232 void hammer_flusher_wait_next(hammer_mount_t hmp);
1233 int hammer_flusher_meta_limit(hammer_mount_t hmp);
1234 int hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1235 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1236 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1237 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1238 int hammer_flusher_haswork(hammer_mount_t hmp);
1241 int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol);
1242 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1243 hammer_volume_t root_volume, int final);
1245 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1246 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1247 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1249 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1250 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1251 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1252 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1253 void hkprintf(const char *ctl, ...);
1254 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1257 int hammer_blocksize(int64_t file_offset);
1258 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1261 * Shortcut for _hammer_checkspace(), used all over the code.
1264 hammer_checkspace(hammer_mount_t hmp, int slop)
1266 return(_hammer_checkspace(hmp, slop, NULL));
1271 static __inline void
1272 hammer_wait_mem_record(hammer_record_t record)
1274 hammer_wait_mem_record_ident(record, "hmmwai");
1277 static __inline void
1278 hammer_lock_ex(struct hammer_lock *lock)
1280 hammer_lock_ex_ident(lock, "hmrlck");
1284 * Indicate that a B-Tree node is being modified.
1286 static __inline void
1287 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1289 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1290 hammer_modify_buffer(trans, node->buffer, NULL, 0);
1293 static __inline void
1294 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1296 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1297 hammer_modify_buffer(trans, node->buffer,
1298 node->ondisk, sizeof(*node->ondisk));
1301 static __inline void
1302 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1303 void *base, int len)
1305 hammer_crc_t *crcptr;
1307 KKASSERT((char *)base >= (char *)node->ondisk &&
1308 (char *)base + len <=
1309 (char *)node->ondisk + sizeof(*node->ondisk));
1310 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1311 hammer_modify_buffer(trans, node->buffer, base, len);
1312 crcptr = &node->ondisk->crc;
1313 hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1314 --node->buffer->io.modify_refs; /* only want one ref */
1318 * Indicate that the specified modifications have been completed.
1320 * Do not try to generate the crc here, it's very expensive to do and a
1321 * sequence of insertions or deletions can result in many calls to this
1322 * function on the same node.
1324 static __inline void
1325 hammer_modify_node_done(hammer_node_t node)
1327 node->flags |= HAMMER_NODE_CRCGOOD;
1328 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1329 node->flags |= HAMMER_NODE_NEEDSCRC;
1330 node->buffer->io.gencrc = 1;
1331 hammer_ref_node(node);
1333 hammer_modify_buffer_done(node->buffer);
1336 #define hammer_modify_volume_field(trans, vol, field) \
1337 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1338 sizeof((vol)->ondisk->field))
1340 #define hammer_modify_node_field(trans, node, field) \
1341 hammer_modify_node(trans, node, &(node)->ondisk->field, \
1342 sizeof((node)->ondisk->field))
1345 * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1346 * created directories for HAMMER version 2 or greater and causes
1347 * directory entries to be placed the inode localization zone in
1348 * the B-Tree instead of the misc zone.
1350 * This greatly improves localization between directory entries and
1353 static __inline u_int32_t
1354 hammer_dir_localization(hammer_inode_t dip)
1356 if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO)
1357 return(HAMMER_LOCALIZE_INODE);
1359 return(HAMMER_LOCALIZE_MISC);